Conceitos essenciais
提案されたCPA-Enhancerは、未知の劣化条件下での物体検出を改善するための革新的なモデルです。
Resumo
Abstract:
Existing object detection methods struggle with unknown degradations.
CPA-Enhancer utilizes CoT prompts for adaptive enhancement.
Demonstrates superior performance in object detection tasks.
Introduction:
Deep learning-based object detection methods have shown promising results.
Strategies involving image restoration and enhancement are common but not always effective.
Current methods are limited to known single degradation scenarios.
Method:
CPA-Enhancer progressively adapts its enhancement strategy under CoT prompts.
Key components include CGM and CPB modules for prompt generation and interaction with input features.
Experiments:
All-in-one setting shows CPA-Enhancer outperforming baseline methods and pre-processing techniques on various datasets.
One-by-one setting demonstrates significant improvements in foggy and low-light conditions compared to existing methods.
Ablation Studies:
Impact of CGM, n, and CPB modules on model performance is analyzed.
Efficiency Analysis:
CPA-Enhancer introduces minimal additional computational cost while achieving superior results.
Estatísticas
CPA Enhancerは、未知の劣化条件下での物体検出において優れた性能を発揮します。